Research On Prediction Of State Of Charge And State Of Heath For Lithium Batteries Based On Machine Learning

Due to the shortage of fossil energy and increasingly serious environmental pollution,new energy electric vehicles occupied an increasing market share in the field of passenger vehicles.Lithium battery has gradually become the main battery of new energy vehicles because of their high energy density,long cycle life,and environmental friendliness.To ensure the safe and efficient use of lithium batteries in new energy electric vehicles,the battery management system needs to monitor and manage them.Among them,SOC and SOH are two important parameters of the battery management system.Their accurate prediction can not only improve the user experience and promote the safe and efficient use of lithium batteries but also provide an important basis for charge and discharge management,balance control,and power prediction of the battery management system.Therefore,combined with the current research status of lithium battery SOC and SOH prediction,and facing the trend of intelligence and networking of battery management systems,this paper designs a more accurate and stable lithium battery SOC and SOH prediction algorithm based on machine learning algorithms.Its main work is as follows:SOC prediction for lithium battery based on attention mechanism and CNN-LSTM fusion model.Aiming at the problems of low prediction accuracy and cumbersome prediction process in the current SOC prediction methods,according to the fact that the charge and discharge of lithium battery is a time process,SOC can be regarded as a time series.Based on recurrent recurrent recurrent neural network,a lithium battery SOC prediction model with a "many to one" structure is designed.The SOC at the current time relates to the discharge data such as voltage,current,and temperature at multiple historical times,and the advantages of the LSTM neural network over RNN and GRU in SOC prediction are verified.Secondly,considering the inconsistent impact of discharge data at different historical times on the current SOC,and making full use of the spatial feature information in the discharge data,this paper combines one-dimensional CNN with LSTM neural network,introduces an attention mechanism,and proposes a lithium battery SOC prediction method based on attention mechanism and CNN-LSTM fusion model.Through ablation experiments and comparative experiments,it is verified that the proposed method has a relatively accurate and stable prediction effect on lithium battery SOC under dynamic discharge conditions.Its average prediction error at different temperatures reaches 0.89%,which is reduced by 81.0%compared with SVM,GRU and XGboost respectively 66.7% and 56.5%,which are better than LSTM and CNN-LSTM.SOH prediction for Lithium battery based on ARIMA-LSTM fusion model.Aiming at the problems of low prediction accuracy and failure to make full use of the effective health characteristics of lithium battery discharge parameters in the current SOH prediction methods,firstly,according to the autocorrelation and linear relationship of lithium battery SOH sequence,this paper uses the differential integrated moving average（ARIMA）model to analyze and predict the SOH sequence,and puts forward the lithium battery SOH prediction method based on ARIMA model.Secondly,according to the degradation mechanism of lithium battery and the factors affecting lithium battery SOH,six key parameters are extracted from the lithium battery discharge data as the health characteristics of lithium battery,and the nonlinear mapping relationship between SOH and health characteristics is obtained based on LSTM neural network.A lithium battery SOH prediction method based on LSTM neural network is proposed.Finally,the predicted values of SOH are combined by linear regression weighting,and the lithium battery SOH prediction method based on the ARIMA-LSTM fusion model is obtained.The first mock exam shows that the prediction method based on the ARIMA-LSTM fusion model has higher prediction accuracy than the single model,and the average absolute error of SOH prediction value is 0.34%,which is better than other commonly used SOH prediction machine learning algorithms.To sum up,this paper have studied the SOC and SOH prediction for lithium batteries,and combined with the shortcomings of its research status,put forward more accurate and stable prediction methods respectively,and lays a solid algorithm for the intelligent battery management system and the lithium battery management platform coordinated by vehiclecloud.The research results have broad application prospects and high application value.